Scribd Logo
Загрузить

Добро пожаловать в Scribd!

  • Kolmogorov

    Загружено:

    Norwegian79
    32 просмотров14 страниц
    Kolmogorov

    Авторское право

    © © All Rights Reserved

    Доступные форматы

    PDF или читайте онлайн в Scribd

    Этот документ был вам полезен?

    Это неприемлемый материал?

    Пожаловаться на этот документ
    Kolmogorov

    Авторское право:

    © All Rights Reserved
    Скачайте в формате PDF или читайте онлайн в Scribd
    Отметить как неприемлемый контент
    32 просмотров14 страниц

    Kolmogorov

    Загружено:

    Norwegian79
    Kolmogorov

    Авторское право:

    © All Rights Reserved
    Скачайте в формате PDF или читайте онлайн в Scribd
    Отметить как неприемлемый контент
    из 14
    228 X. PRIME NUMBERS Suggested Reading ory of numbers, ‘he theory of numbers, ‘theory, Graylock Press, Rochester, ar, MeGrow-Hil, New York, 18 ‘he theory of numbers, Perganion Press, Ur Mathematics les Combet, MeHMeods a 7) THE THEORY OF PROBABILITY A.M kOLH oe soxncVy SL, The Laws of Probability n event 4, which ‘complex of con land sometimes does not occur, is called random with respect to the complex o raises the question: Does the randomness of the event 4 demonstrate the absenee of any law connecting the complex of conditions and the event A? For exampl be established that amps of a specific type, manufactured in a certain factory (condition $) sometimes continue to burn more than 2,000 tours (event ), but some- times burn out and become useless before the ex May it not still be possible that the results of exper 230 XI. THE THEORY OF PROBABILITY Or should we restrict ourselves to indicating For which in practice al ‘eases the lamps work for evaluation of the gu will of showing for any T the percent (7) of the lamps which work for no Iess than 7 hours, say in the form of ‘the graph in figure 1, ‘The curve »(7)is found in practice by testing with a sufficiently large sample (400-200) of the lamps. Of course, the curve ° is only if the Some ep je snl wil ive approxinatly the sa (ie, the new curve (7) will differ 2 the ist a a7) is given by a function (7), where a single lamp (made u §2. THE AXIOMS AND BASIC FORMULAS 231 This example belongs to the laws ‘The reality of suck laws eannot be ‘What was said about the “closeness” of the frequency » to the prob- ability p for a large number n of tests is somewhat vague; we may be for any n. T . The very statement indefiniteness in character, hat and p are clase t0 © as becomes cledr if' we try to make the whole situation precise §2. The Axioms and Basic Formulas of the Elementary Theory of Probability roceeding in a purely empirical way. Since « law of is natural to imagine ‘@ mass experiment. tight, As soon as we have ‘we may proceed to a 2 32 XL. THE THEORY OF PROBABILITY Probability PUA) of any event A must be assumed to fie between zero O< P<], An event Aisle te mion octets Avandia eae, one of the event dys Fot example ie int Or ad consis er of coca el ya thee anon mm ei fr te cree of two tually exchsive events A= AU Ay, we iH wve the 9 sponding frequencies » = vy = % ‘This leads naturally to the follo wat lly t0 the following axiom for the addition of proba ayy Ad READ, a re events Ay and A, Further, for an ev fe and A, UA, denotes their Pu) = 1 o theory of probability is constructed on the {ype (D, (2), and G). Fram the point of probability is a numerical Fan 1e properties of serve as a sufficient ntary theory of prob- expressed by Formula (1, Q) nd basis forthe eonsrution of what sealed the ees ably, iF we do not ling ot ion of events a8 defined eit is more useful oo afte ut to relics fm practic life cannot be abe the compete Terma ly mathematical presentation of the Sys Ay the event A consintng of the oeutrence of at eat one ofthese mente ie of these events, For breviy we ge PLAS) (0 PUA} §2. THE AXIOMS AND BASIC FORMULAS 233 number of pairwise ion A, m ofaddition, we easily obtain for a ive events 4,. Ay. A. and their (4) = PEA) + PAR) bo + PAD (the so-called theorem of the addition of probabitites). If the union of these events is an event that is eertain complex of conditions S one of the events 4, ntust osc . under the then P(A) + PCA) + + PCA) = 1 In this ease the system of events A, ,°, A, is called a complete system of events ‘We now consider two events 4, and B, which, generally speaking, are y exclusive, The event Cs the intersection of the events 4 and C_~ AB, if the event C consists of the oscurrence of both wz an even number in the For example, the event 4 consists of obi ple of three, then the throw of a dic and B consists of obtaining a mt eventt C consists of obtaining a six. Ina large number 1 of repeated trials, let the event 4 occ: nes and the event B occur times, ink of which 2 occurs together with the event 4, ‘The quotient k/nt is called the conditional frequency of the event B under jon A, The Frequencies and kip are connected by the the core formula ly P(BJA) of the event B under the condition Ais the quotient PUB) PIA) = “pray Here itis assumed, of course, that P(A) # 0. If the events A and # ate in no way essentially connected with each other, then it is natural 19 assume that event £# will not appear more often, for less often, when A has occurred than when A has sot oecurred, ie hat approximate I XI. THE THEORY OF PROBABILITY (eequation frequency of in ~ », is the frequency of the eves we event B and finally kin — events 4 and 2. nist For the protien ofthe cents A,B and A ist aceept the correspor ‘exact equati " PAB) = PCA) PCB) @ (4) serves to define the dndependence of two events A and B. larly, we may define the independence of any number of events, We may sive a definition of the independence of any nuinber of experiments, which means, roughly speaking, that the outcome of any part of o not depend on the outcome of the rest,? pute the probability P, of precisely & occurrences of a ndependent tests in each one of Which the probability ame. We denote by 4 the event is obvious that PEA) 1 P(A) 1 =p. 1¢ definition of the independence of experiments itis easy to probability of any specific sequence consisting of & occurrences ‘ nonoveucrenees af A is ail — pe, © 5; for example, for n ~ 5 and k = 2 the probability of robability of getting the sequence 44.444 will be p(l — p) {1 — p)(1 —p) = pil — pF By the theorem on the addition of probabilities, P, will be equal to the sum of the probabilities of all sequences with & occurrences and event 4, i.e. Py will be equal from (5) to the product ‘amber of such sequences by pY(l — p)'-", The number of such A more exset meaning of independent exp periments tn any ay preasgrel outcome, and the ever have pease comes, for arirarydecompeation ito 180 events ain! Bae independent inthe ese We wl return in oa consideration of he edie (of the indepensience of evens. va actual world 2. THE AXIOMS AND BASIC FORMULAS 235 sequences is obviously equal to the number of combinations of 1 things taken k at fouleomes may occupy any k places sequence of m trials. Finally we get Pym Chel — py® = 0.1,2,09) © (which is called a binomial distribution) nsand formulas are applied, we consider ory of gunfire tution of the target, What interests si to assume that 40 shots will insure the necessary five hits. A purely empirical solution of the problem would proceed as follows, For given dimensions of the target and for @ siven range, we earry out a large number (say 200) of firings, each con- sisting of 40 shi iny of these firings produce In order to see how the definit at least five i ple, by 195 firings out of the 200, then the probability Pis approximately equal to 195 97 = YE ons. purely empirical way, we will use up 8,000 si If we proceed in t by assuming that the lependent of the size of the 1 the Tongitudinal and lateral deviations, from the shells, follow a law with respect to the of various sizes strated in figure 2. scattesing of the she! target. It turns out ‘mean point of landing Frequency of deviat 2% 7% 16% 25% 25% 16% 7% 2% 42-38 28-8 0 8 28 38 45 | Fic. 2 B here denotes what is called the probable deviation. The probable deviation, generally speaking, is different for longitudinal and for lateral deviations and increases with increasing range. The probable ions for different ranges for each type of gun and of shell empirically in fring practice on an artillery range. But the solution of all possible special problems of the kind deserited is carried ‘out by calculations only. For simplicity, we assume target has the form of a rectangle, 236 XI. THE THEORY OF PROBABILITY ine of fire and has a length of two +, while the other side is perpendicular 10 © probable lateral deviations. has already been well established, 50 ls passes through its center (gure 2), jons are inde- is necessary and B99 —b ay ayer Tank f getting no less than five hits, and is now expressed by the formula 92. THE AXIOMS AND BASIC FORMUI 237 9 - B EE (00113, ad Q= 0016, P= 0.984. so blaine is somewhat lose 10 ie sufficient in the theory of gunfire. Most ofte: umber of shells ned mp 238 XL. THE THEORY OF PROBABILITY considerations of sme. For example th uta onsason ‘a die, Le.. a cube made of a homogeneous material wi ‘h ‘ a toa safer hight, with equal probably on 6 raced lng before there wat ny stemaictecunlaton att ‘enly it by abseration Systematic experiments af tit ind ae tare Cama out in he ast thes semi, ciety by auton of eto they of probe when the tony of probate Screoped aden, The results of tha epoca ven Bat the gestion of extening them 10 aealopons ee sstely arouses interes. Foretapl far a6 KooM, nO ON hoe canned out suo etna sxpernets nts omoqenco ee wih othe se Gt tise no. dea at we wee mae 2.00 sich tsi, the teesded le woul show tach of Se ee approximately a thousand times. Show sash of The bo pobab sensi so ay Samp in ra cs derived from arguments of symmetry or homoe ai ts Peale ma Spence motion in 8 gas; another cast whee the susie he Saye Be atin. of tasley Of coins bnaetioae sess we pthc uth snp oman 3. The Law of Large Numbers and Limit Theorems Itis completely natural to wish for greater quantitative precision in the hat in a “long” series of tests the frequency of an occurrence: probability. But here we must form a clear notion the problem. In the most typical eases in the isseries either of the he Frequency stel and Hono, For exampl tials, te probability that we will tam un a six o Kars O §. LARGE NUMBERS AND LIMIT THEOREMS 20 in other words, with probability (4)° we will obtain a frequency of rolling 4 six which is equal to ane; and with probability (1 — 3)" > Oa six will rot come up at all, ie., the frequency of ix will be equal to zero. In all similar problems any nontrivial estimate of the closeness of the Frequency to the probability cannot be made with complete certainty, but only with some probability less than one. For example, it may shown that in independent tests,” with constant probability p of the ‘occurrence of an event in each test the inequality p| < 0.02 o will be satisfied, for m = 10,000 (and any p), with, Tor the frequency 1 probability P > 09999, @ Here we wish first ofall to exaphesize that in this formulation the quanti- tative estimate of the closeness of the frequency jn to the probability p involves the introduction of a new probability P. ‘The practical meaning of the estimate (8) is this: If we carry out of m tests each, and count the M sets in which inequality (7) is sal sn for sufficiently large WV we wil have approximately M MA. p> 0.5999. 0 Am P > 039998, O) But if we wish to define the selation (9) more precisely, either with respect to the degree of closeness of M/N to P, ot with respect to the confidence with which we may assert that (9) will be verified, then we must have recourse to general considerations of the kind introduced previously in discussing what is meant by the closeness of p/n and p. Such considera- tions may be repeated as often as we like, but itis elear that this procedure will never allow us to be froc of the necessity, at the last stage, of referring to probabilities in the primitive imprecise sense of this term. Te would be quite wrong to think that in some way to the theary of probability. In a mathematical investigati of actual events, we always make a model of them. ‘The discrepancies hetween the actual course of events and the Uheoretiesl model can, in its turn, be made the subject of mathematical investigation. But for these discrepancies we must construct a model that we wall use without formal ‘mathematical analysis of the discrepancies which again would arise in it in actual experiment The proof of the estimate (8) 6 ciscussed Laer inthis secfon SS 240 XI. THE THEORY OF PROBABILITY We note, moreover, that in an actual pplication of the estimate® P{)4—p| < 0.02] > 09999 Oo) {0 one series of » tests we ure already depending on certain considerations of symmetry: ineq ows that for a very large number N of series of tests, relation (7) will be satisfied in no less than 99.09% of t iow it is natural to expect with great confidence that inequality in particular to that one of the sequence of m tests which is lerest to us, but we may expect this only if we have some reason for assuming that the position of this sequence among the others is regular special feauures. s thal we may decide to neglect are different in different jons. We noted earlier that our preliminary enleulatious for expenditure of shells necessary to produce a given result meet the is to be solved with probability 0.95, ie, the = do not exceed 0.05. This standard is expla if we were to make calculations neglecting a probat iat in practice we would conclude that the in the time at our disposal, or with the task cou given supply of shells, basis of nel be adopted preparation is proposed, and to tes es we plan to use it in {en cases, chosen without bias from among the patients suffering from liness. Here we agree that the advantage of the new preparation wi bbe considered as proved iF it gives « positive resul cases oul of the ten. This easy to calculate der of 0.05 of getting a wrong res the new preparation when in fact itis ly effective or even worse than the old, For if in each of the ten. Ws, the probability of a positive outeome is equal to p, then the “This the aexepied notation for estimate (8) of the proba of ine vO, §8. LARGE NUMBERS AND LIMIT THEOREMS 241 probability of obtaining in fen experiments 10, 9, or 8 positive outcomes, is equal respectively to Py =P Py = 1PM — ph, Py = 450% — a. For the case p — the sum of these is 56 a = Be a ol P= Pat Pat Pa = igqq ~ 098. latin fact the new preparation is ith probability of order 0.05 the better than the old. To reduce the number of experi= his way, under the assu eas the old, we eed to agree se va wearin rogttomont seems foo severe tothe advoctes ofthe now rejaeton it wl be aeesry 10 make the number of exetnents rreefraby larger than TO. For example, form = TOD. itwe ages that the advantage of the new preparation is proved far n> 65, then the ty be P= DOTS Wetgaons osu of 005 is eka oes the laws of probability are hey involve the negleet of smaller probabilities. We vill return to this question later. . In the previous examples, we have made use of particular cases of the binomial formula (6) Pu = Ch for the probability of getting exac trials, in each one of which a pk where yc is the actual number of positive results.” Ot n ith probability Pas ie = Pas Here ye takes the values m — SS 22 Xl. THE THE THEORY OF PROBABILITY @. LARGE NUMBERS AND LIMIT THEOREMS 43 the sum of those P,, for whic! n satisfies the F(T) converge to one, timate of the probability —p| 0, wve been constructed, Here is & ion F(Z) del -xeerpt feom them, riot | 2 3 F 0.68269 [0.93450 10.9730 tables SL 2a XI. THE THEORY OF PROBABILITY @. LARGE NUMBERS AND LIMIT THEOREMS 245 ng of variance and standard which gives complete cer en Itshows that devintions of & from M(G)signcantly greater than 9 are [AS for the sum of random variables ib fi foneh always satisfy the eq dm thematical expect M(E) = Mt = PE = 3). for all possible values of the variable ¢ is DO =D Bis, — ME re square root of the variance i called the standard devia nL 246 XI. THE THEORY OF PROBABILITY We now assume that for each of these terms the vari exceed a certain constant He arinse does ot DEN

    Вам также может понравиться